Orthogonal Nonnegative Tucker Decomposition

In many data analysis problems, the columns of A are corresponding to data points, for instance, images of pixel intensities. NMF has been successfully applied into many fields including image processing, text data mining and so on. It has been demonstrated that NMF is a powerful technique for dimension reduction. Compared to other well-known method, like singular value decomposition or principal component analysis, NMF is able to give more interpretable results due to its combinations of nonnegative basic vectors. In general, NMF is NPhard and the solution is not unique. It is necessary to impose additional constraints on the factor matrix like orthogonality constraints.